Narrative framing shifts LLM cooperation by 0.307 at identical payoffs.

A July 22 paper defines strategic robustness as behavioral invariance across payoff-equivalent framings, measured over 24 model-game-context cells and 7,200 decisions across four social-dilemma games (arXiv 2607.19670). Pooled robustness was 0.783, with friend-sharing framing raising cooperation by 0.307 versus business framing despite identical action sets and payoffs. Models tested are old (GPT-3.5, GPT-4, LLaMa-2) via published cooperation rates, which limits any claim about current frontier agents. The method transfers though: evaluate with multiple payoff-equivalent phrasings, not one.